design of a low power, high performance track-and …18676/fulltext01.pdf · design of a low power,...

Design of a Low Power, High PerformanceTrack-and-Hold Circuit in a 0.18µm

CMOS Technology

Erik Säll

LiTH-ISY-EX-3248-2002SEPTEMBER 3, 2002

Design of a Low Power, High PerformanceTrack-and-Hold Circuit in a 0.18µm

CMOS Technology

Master Thesis

Division of Electronic Devices

Department of Electrical Engineering

Linköping University, Sweden

By

Erik Säll

LiTH-ISY-EX-3248-2002

Supervisors: Niclas Hallqwist (Acreo AB)Patrik Eriksson (Acreo AB)

Examiner: Prof. Christer Svensson

Linköping, September 3, 2002

Avdelning, InstitutionDivision, Department

Institutionen för Systemteknik581 83 LINKÖPING

DatumDate2002-09-03

ISBN

ISRN LITH-ISY-EX-3248-2002

Serietitel och serienummerTitle of series, numbering

ISSN

SpråkLanguage

Svenska/SwedishX Engelska/English

RapporttypReport category

LicentiatavhandlingX Examensarbete

C-uppsatsD-uppsatsÖvrig rapport____

URL för elektronisk versionhttp://www.ep.liu.se/exjobb/isy/2002/3248/

TitelTitle

Design av en lågeffekts högprestanda track-and-hold krets i en 0.18µm CMOSteknologi.

Design of a Low Power, High Performance Track-and-Hold Circuit in a 0.18µm CMOSTechnology

FörfattareAuthor

Erik Säll

SammanfattningAbstractThis master thesis describes the design of a track-and-hold (T&H) circuit with 10bit resolution,80MS/s and 30MHz bandwidth. It is designed in a 0.18µm CMOS process with a supply voltage of1.8 Volt. The circuit is supposed to work together with a 10bit pipelined analog to digital converter.

A switched capacitor topology is used for the T&H circuit and the amplifier is a folded cascodeOTA with regulated cascode. The switches used are of transmission gate type.

The thesis presents the design decisions, design phase and the theory needed to understand thedesign decisions and the considerations in the design phase.

The results are based on circuit level SPICE simulations in Cadence with foundry provided BSIM3transistor models. They show that the circuit has 10bit resolution and 7.6mW power consumption,for the worst-case frequency of 30MHz. The requirements on the dynamic performance are allfulfilled, most of them with large margins.

NyckelordKeywordTrack-and-hold, CMOS, 0.18, low power, high performance, 10-bit, folded cascode, switch theory,correlated double sampling, CDS, fully differential, gain boosting, regulated cascode, transmissiongate, transmission gate switch, clock generator, clock driver, bias, bias circuit, amplifier design,switch design, common mode feedback, CMFB, 80MSPS, 80MS/s

ABSTRACT

This master thesis describes the design of a track-and-hold (T&H) circuitwith 10bit resolution, 80MS/s and 30MHz bandwidth. It is designed in a0.18µm CMOS process with a supply voltage of 1.8 Volt. The circuit issupposed to work together with a 10bit pipelined analog to digital converter.

A switched capacitor topology is used for the T&H circuit and the amplifieris a folded cascode OTA with regulated cascode. The switches used are oftransmission gate type.

The thesis presents the design decisions, design phase and the theory neededto understand the design decisions and the considerations in the designphase.

The results are based on circuit level SPICE simulations in Cadence withfoundry provided BSIM3 transistor models. They show that the circuit has10bit resolution and 7.6mW power consumption, for the worst-casefrequency of 30MHz. The requirements of the dynamic performance are allfulfilled, most of them with large margins.

ACKNOWLEDGMENTS

I would like to take the opportunity to thank the people at Acreo AB inNorrköping, especially them at the System Integration department, for theirsupport and for making this time to a pleasant and memorable ending of mystudies.

I would especially like to thank Patrik Eriksson and Niclas Hallqwist forgiving me the opportunity of this interesting work and Niclas for hisguidance during the work. I would also like to thank Håkan Träff for veryvaluable discussions on circuit architectural matters during the master thesiswork and Maria Hofvendahl (my “desk-neighbour”) for her support duringthe work.

Finally I would like to thank my family and friends (all included) for all theirsupport, not only during this work, but also during the past five years atLinköping University. Thank you all!

Erik Säll

CONTENTS

1 INTRODUCTION ................................................................................ 1

1.1 PURPOSE.............................................................................................. 1

1.2 READING GUIDELINES .......................................................................... 1

1.3 ABBREVIATIONS................................................................................... 2

2 SPECIFICATION ................................................................................ 4

2.1 FS ....................................................................................................... 5

2.2 SNR .................................................................................................... 5

2.3 SFDR .................................................................................................. 6

2.4 THD.................................................................................................... 6

2.5 SNDR ................................................................................................. 6

2.6 TWO-TONE IMD .................................................................................. 7

3 THEORY .............................................................................................. 8

3.1 BASIC MOSFET THEORY .................................................................... 8

3.2 DEFINITION OF SOME MOSFET PARAMETERS....................................... 9

3.3 PARASITICS........................................................................................ 11

3.4 NOISE IN MOSFET ............................................................................ 12

3.4.1 Flicker Noise (1/f noise) ..............................................................13

3.4.2 Thermal Noise .............................................................................13

3.5 CURRENT MIRRORS............................................................................ 13

3.6 FEEDBACK ......................................................................................... 15

3.6.1 Settling Time................................................................................15

3.6.2 Linear Settling .............................................................................17

3.7 MINIMUM THEORETICAL POWER CONSUMPTION FOR T&H ................. 18

3.8 COMMON MODE FEEDBACK (CMFB) ................................................. 19

3.9 DIFFERENT A/D STRUCTURES ............................................................ 20

3.9.1 Flash Converter...........................................................................20

3.9.2 Successive Approximation Converter ..........................................21

3.9.3 Algorithmic Converter.................................................................21

3.9.4 Pipelined Converter ....................................................................22

3.10 SWITCH CHARACTERISTICS OF MOSFET ........................................ 23

3.10.1 Charge Injection..........................................................................23

3.10.2 Dummy Transistors .....................................................................25

3.10.3 Bottom Plate Sampling ................................................................25

3.10.4 Clock Feed Through....................................................................26

3.11 OFFSET CANCELLATION.................................................................. 26

4 BRIEF PRESENTATION OF DIFFERENT STRUCTURES......... 29

4.1 TRACK-AND-HOLD............................................................................. 29

4.1.1 Why Fully Differential? ...............................................................29

4.1.2 Open- and Closed-Loop...............................................................29

4.1.3 Switched-Capacitor .....................................................................30

4.1.4 Switched-Current ........................................................................30

4.2 OPERATIONAL TRANSCONDUCTANCE AMPLIFIER (OTA)..................... 30

4.2.1 Single-Stage ................................................................................31

4.2.2 Two-Stage ...................................................................................31

4.2.3 Telescopic Cascode .....................................................................32

4.2.4 Regulated Cascode (Gain Boosting)............................................33

4.2.5 Folded Cascode...........................................................................34

4.3 SWITCHES .......................................................................................... 35

4.3.1 Single MOS Switch ......................................................................35

4.3.2 Transmission Gate Switch ...........................................................36

4.3.3 Bootstrapped Switch ....................................................................37

4.4 CLOCK GENERATOR AND CLOCK DRIVER ........................................... 38

5 MOTIVATION OF SELECTED STRUCTURE.............................. 39

5.1 TRACK-AND-HOLD............................................................................. 39

5.2 OTA.................................................................................................. 40

5.2.1 Why NMOS or PMOS at Input-Stage?.........................................41

5.2.2 Bias Circuit for the OTA..............................................................42

5.3 SWITCH.............................................................................................. 43

5.4 CLOCK GENERATOR ........................................................................... 43

5.5 CLOCK DRIVER .................................................................................. 45

6 DESIGN .............................................................................................. 46

6.1 OPERATIONAL TRANSCONDUCTANCE AMPLIFIER (OTA)..................... 47

6.1.1 Finding Specification for the Amplifier........................................47

6.1.2 Calculation of Lower Limit Values ..............................................52

6.1.3 Vgs of Input Transistors ..............................................................55

6.1.4 Input and Output DC Levels ........................................................55

6.1.5 Current Mirror Transistor Sizes ..................................................55

6.1.6 Approach for Simulations ............................................................56

6.1.7 Design of Gain Boost Amplifiers .................................................57

6.1.8 Design of Common Mode Feedback (CMFB) ..............................59

6.2 BIAS CIRCUITRY ................................................................................ 60

6.2.1 Calculation of Approximate Values .............................................60

6.2.2 Design and Simulation Flow........................................................60

6.3 SWITCH ARRANGEMENT..................................................................... 62

6.4 ON-CHIP CLOCK DRIVER AND CLOCK GENERATOR............................. 67

7 RESULTS ........................................................................................... 72

8 CONCLUSIONS................................................................................. 74

8.1 DISCUSSION ....................................................................................... 74

8.2 FUTURE IMPROVEMENTS .................................................................... 75

APPENDIX I – DERIVATION OF TOTAL OUTPUT NOISE ............ 76

APPENDIX II – DERIVATION OF DEPENDENCY TABLE ............. 79

REFERENCES ......................................................................................... 81

1 - INTRODUCTION

1

1 INTRODUCTION

The track-and-hold (T&H) circuit is a fundamental block for analog-to-digital (A/D) converters. Its use allows most dynamic errors of A/Dconverters to be reduced, especially those showing up when using highfrequency input signals.

1.1 Purpose

The purpose with this project was to design a front-end T&H, to be usedtogether with a 10bit pipelined A/D converter. It was to be designed in a0.18µm CMOS process, which only allows a supply voltage of 1.8Volt.

The main design consideration was to minimize the power consumption,since the circuit is supposed to be used in a mobile transceiver. Low powerconsumption is thereby crucial.

1.2 Reading Guidelines

This master thesis starts by presenting the theoretical background needed tounderstand the ideas and different structures presented in the chapters thatfollow. It is also important for the understanding of how and why somedesign considerations are done in chapter 6.

At the end, the results as well as the conclusions are presented. The resultsare based on circuit level SPICE simulations in Cadence, with foundryprovided BSIM3 transistor models. Some suggestions for furtherimprovements and alternative design approaches are also presented.

A brief summary of all the chapters is found below.

Chapter 2. Specification

Presents the specification of the T&H and explains the definitions ofsome of the most important parameters.

Chapter 3. Theory

The theory needed for the chapters that follow are briefly presented.

1 - INTRODUCTION

2Chapter 4. Brief Presentation of Different Structures

Some of the most commonly used structures for both track-and-holdsas well as their building blocks are introduced.

Chapter 5. Motivation of Selected Structure

The structures chosen among the ones presented in chapter 4 are morethoroughly explained and motivated.

Chapter 6. Design

The different steps of the design phase are presented and explained.Some derivations of the demands on the performances of the differentparts are also performed.

Chapter 7. Results

The simulation results are presented and commented on.

Chapter 8. Conclusions

Some conclusions made during the design are presented, as well assome suggestions for future improvements.

Appendix I – Derivation of Total Output Noise

Derivation of the total output noise of the track-and-hold circuit.

Appendix II – Derivation of Dependency Table

Derivation of the dependency table, which is used during the design ofthe amplifier.

References

Only 14 of the references are referenced to in the text. The others areused during the literature study phase in the beginning of this work.

1.3 Abbreviations

• T&H - Track-and-Hold

• A/D - Analog-to-Digital converter

• SNR - Signal-to-Noise Ratio

1 - INTRODUCTION

3

• SFDR - Spurious Free Dynamic Range

• SNDR - Signal-to-Noise-and-Distortion Ratio

• IMD - Intermodulation

• FS - Full Scale

• THD - Total Harmonic Distortion

• MOSFET - Metal Oxide Semiconductor Field Effect Transistor

• NMOS/PMOS - N-channel MOSFET/P-channel MOSFET

• OP/OP-amp - Operational Amplifier

• OTA - Operational Transconductance Amplifier

• CMFB - Common Mode Feedback

• CDS - Correlated Double Sampling

• RMS - Root Mean Square

• SR - Slew Rate

2 - SPECIFICATION

4

2 SPECIFICATION

The specification of the track-and-hold (T&H) can be found in the tablebelow. The most important parameters are explained in the chapters thatfollow.

The signal swings were not specified prior to this work; instead they werespecified during the design phase.

Table 2-1 Specification for the Track-and-Hold.

Sampling Frequency and Input Signal Frequency RangeParameter Frequency Range CommentSampling

FrequencyRange

10 to 80 MS/s

Input SignalFrequency

RangeDC to 30 MHz

Specification of Input, Clock and OutputParameter Signal Swing Type Comment

Input 600 mV (Vpp) Differential Defined as FSClock 3.6 V (Vpp) Sinusoidal, differential

Output 600 mV (Vpp) Differential

Circuit Parameter SpecificationParameter Requirement Comment

Voltage Gain 0 dB (Nominal)Unity gain buffer.

DC- 30 MHzOutput DC-

offset < 50 µVCorrelated doublesampling assumed

Hold-modefeedthrough < -65 dB

Worst case DC - 30MHz

Dynamic PerformanceParameter Signal Level Signal

Frequency Requirement Comment

SNR -1 dBFS DC – 30 MHz 66 dBThermal and

(internal) clockinduced noise

SFDR -1 dBFS DC – 30 MHz 64 dB Highest spuriousSNDR -1 dBFS DC – 30 MHz 61 dB 10 effective bits

Two-toneIMD -7 dBFS DC – 30 MHz 58 dB -7 dBFS for each tone

2 - SPECIFICATION

5

2.1 FS

FS stands for “Full Scale”, and is the same as the peak-to-peak voltage (Vpp)of the differential signal, or twice the Vpp of one of the input signals (seeFigure 2-1). Therefore, a signal level of –1 dBFS means that the signal levelis 0.89 of the maximum FS signal level.

Figure 2-1 Definition of Full Scale (FS) [1].

2.2 SNR

Signal-to-Noise Ratio (SNR) is defined as:

[ ]dBpowerNoise

powerSignallog1010SNR

×=

Maximum SNR is equal to the ratio of the maximum sinusoidal power to thequantization noise. If oversampling is taken into account, the maximumachievable SNR for an A/D converter becomes:

( )[ ]dBOSRlog101076.102.6NmaxSNR ×++×=

OSR = Oversampling Ratio =0

S

f2

f, where f0, fS are the signal bandwidth and

sampling frequency respectively and N is the number of bits.

2 - SPECIFICATION

6

2.3 SFDR

Spurious Free Dynamic Range (SFDR) is defined as the ratio between themaximum signal component to the largest distortion component [2].

componentdistortionLargestcomponentsignalMaximum

SFDR =

2.4 THD

The total harmonic distortion (THD) of a signal is defined as the ratio of thepower of the second and higher order harmonics to the power of thefundamental for the signal.

+++=2f

24

23

22

V

VVVlog10THD

L

2.5 SNDR

Signal-to-Noise-and-Distortion Ratio (SNDR) is defined as:

[ ]dBpowerDistortionpowerNoise

powerSignallog10SNDR 10

+

×=

The maximum SNDR that can be reached with an A/D converter of N bits is(sine-shaped input signal):

[ ]dBTHD1.766.02NSNDR max −+×=

If the A/D is oversampled the maximal SNDR becomes:

( ) [ ]dBTHDOSRlog101.766.02NSNDR 10ngoversamplimax, −×++×=

Since we want a T&H that delivers an output signal with at least 10bitresolution, the SNDR of the T&H must be equal or larger than SNDRmax. Onthe other hand, since oversampling is used in the A/D converter, the demandon SNDR can be relaxed with a magnitude equal to the oversampling factorbelow.

( ) [ ]dBOSRlog10 10×

2 - SPECIFICATION

7

2.6 Two-Tone IMD

Two-tone intermodulation (IMD) is defined as the ratio between the signal-power (two different frequencies) to the power of the third orderintermodulation tones ( 21 ff2 − and 12 ff2 − ).

Figure 2-2 Definition of two-tone intermodulation.

2f1-f2 f1 f2 2f2-f1

IMD

Sign

alm

agni

tude

3 - THEORY

8

3 THEORY

This chapter treats the basic theory needed to understand for example howand why certain design considerations are done. The purpose is also to putthe track-and-hold circuit into a context, and from that be able to understandits task and how different parameters affect the performance of the T&Hitself as well as the surrounding. The emphasis is on the results, not on thederivations. For those who are interested, the derivations can be found in [3],[4], [5] or [6] if nothing else is said.

3.1 Basic MOSFET Theory

The figure below shows the important dimensions of an NMOS transistorand the corresponding schematic symbol.

Figure 3-1 Cross section of NMOS transistor showing the important dimensionparameters and the corresponding circuit symbol [4].

3 - THEORY

9

3.2 Definition of some MOSFET parameters

The different regions of operation are shown in the figure below, where theI-V characteristics of a transistor are shown. Note that the triode region andthe active region are equivalent to linear region and saturation regionrespectively.

Figure 3-2 Regions of operation for NMOS transistor [4].

The expressions for the drain current in each region are as follows for anNMOS- and a PMOS-transistor respectively (short channel effects notconsidered).

NMOS:

( )

( ) ( ) DStnGSDS2

tnGSoxnD

tnGSDS

2DS

DStnGSoxnD

tnGSD

VVV0;V1VV2L

WCI

VVV0;2

VVVV

LW

CI

VV;0I

<−<λ+−µ=

−<<

−−µ=

<=

3 - THEORY

10

PMOS:

( )

( ) ( ) SDtpSGSD2

tpSGoxpD

tpSGSD

2SD

SDtpSGoxpD

tpSGD

VVV0;V1VV2L

WCI

VVV0;2

VVVV

L

WCI

VV;0I

<−<λ+−µ=

−<<

−−µ=

<=

Where Cox is

ox

0rox t

Cεε=

By using the equations above, the following expressions can be derived byusing the definitions of gm, gmbs and gds [4].

( )( ) Dox0dsTgsox0gs

Dm I

LW

C2V1VVLW

CV

Ig µ≈≈λ+−µ=

∂∂≡ K

mbsF

mmbs

gs

Dm

bs

T

gs

D

bs

gs

gs

D

bs

Dmbs

gV22

gg

V

Ig

V

V

V

I

V

V

V

I

V

Ig

η=−Φ

γ=

⇒⇒∂∂==

∂∂−

∂∂=

∂∂

∂∂=

∂∂≡ K

3 - THEORY

11

( ) D2

Tgsox0ds

Dds IVV

L

WC

V

Ig λ≈−λµ=

∂∂≡

WhereL

1∝λ

3.3 Parasitics

The parasitics of a MOSFET can be modelled in several ways with differentdegrees of complexity. Which model to chose depends on the applicationand in what stage of the design you are. One common way of modelling thesmall signal parasitic capacitances is shown in Figure 3-2. The complexity ison a reasonable level for being used in the earlier stage of the design, wherecalculations by hand are used to get a grip of for example what performanceto expect.

Figure 3-2 Cross section of an NMOS transistor with parasitic capacitances [4].

The equations for the parasitics can be found in Table 3-1. The derivationsare not shown, but can be found in [7].

3 - THEORY

12

Linear region

=φ

2i

DAB0

n

NNln

q

Tk NA= Acceptor concentrationND= Donor concentration ni=Intrinsic concentration

oxovoxgsgd CWLWLC2

1CC +==

0

sb

s0j

dbsbV

1

2

WLAC

CC

φ+

+

==

Saturation regionPs, Pd =

Source and drain perimeterAs, Ad =

Source and drain areaCj0, Cj-sw0 = constant

oxovoxgs CWLWLC3

2C += oxovgd CWLC =

( ) swjsjss

swssbsb

CPCWLA

CCC

−

−++=

=+′=

swjdjdd

swddbdb

CPCA

CCC

−

−+=

=+′=

0

SB

0jjs

V1

CC

φ+

=

0

SB

0swjswj

V1

CC

φ+

= −−

0

DB

0jjd

V1

CC

φ+

=

Table 3-1 Equations for calculating parasitic capacitances in linear- andsaturation-region.

3.4 Noise in MOSFET

Flicker noise ( f1 -noise) and thermal noise are the dominant noise sources inMOSFET transistors working in the active region. That is why the noisemodel often used is the one in the figure below.

Figure 3-3 Noise model for MOSFET in saturated (active) region [4].

3 - THEORY

13

3.4.1 Flicker Noise (1/f noise)The flicker noise is a result of vacant trap levels for the majority carriers[7][8]. This will result in trapping of some of the majority carriers in arandom manor. From the world outside this appears as noise. The flickernoise is modelled as a frequency dependent voltage source in series with thegate, of value

fWLC

K)f(V

ox

2g =

Where K is dependent on the characteristics of the device, it can thereforevary a lot between different processes.

Worth noting is that a PMOS transistor has got lower flicker noise since itsmajority carriers (holes) are less likely to be trapped.

3.4.2 Thermal NoiseFor a transistor in the triode region the thermal noise is simply due to thechannel resistance giving a thermal noise current in the drain. The noisecurrent is then given by:

ds

B2d r

Tk4)f(I =

When the transistor is in the active region the channel cannot be consideredhomogenous. The resistance giving rise to the thermal noise must then bederived by integrating over small portions of the channel. The resultingexpression for the noise current then becomes:

mB2d g

3

2Tk4)f(I

=

For the case where VDS=VGS-Vt.

3.5 Current Mirrors

To bias an amplifier (set the desired bias currents) one would like to use acurrent source that is as close to an ideal current source as possible. Thismeans that the current source should have high output impedance.

3 - THEORY

14

To realize a current source, current mirrors are often used and to bias anamplifier, current mirrors are almost always used. According to above, thecurrent mirror should be designed to have as high output impedance aspossible. How can this be done?

Take a look at the cascode current mirror and its equivalent small signalcircuit below.

Figure 3-4 Cascode current mirror and its equivalent small signal circuit.

From the equivalent small signal circuit above the output resistance isderived below:

(a) ( ) 4dsx2dsxout2mx gVgVVgV =−+−

(b) ( ) 2dsx2dsout2mx2dsxout2mxout gVgVgVgVVgVI −+−=−+−=

(a), (b) => ⇒≈⟩>>⟨≈++

=2m

4ds2dsoutdsm

4ds2ds2m

4ds2dsoutout g

ggVgg

ggg

ggVI

4ds2ds

2mout

out

outgg

gR

I

V ==

From above we have that the output impedance is equal to

3 - THEORY

15

224d4ds2ds

2mout LWLconstantkeptI

gg

gR ∝⟩⟨∝=

To maximize the output impedance, large lengths and maximum W2 shouldbe used.

3.6 Feedback

Figure 3-5 Negative feedback amplifier block diagram [3].

Feedback is often used to linearize and/or increase the bandwidth ofamplifiers. The output noise referred to the amplifier (“a” in Figure 3-5) canalso be reduced in some cases [9]. Feedback can also be used to make theamplifier unstable and thereby creating an oscillator. In this design feedbackis used to create a unity gain buffer and to be able to perform doublecorrelated sampling.

3.6.1 Settling TimeThe settling time is the time it takes for the signal to settle within a certainwanted range [3]. It is illustrated in the figure below, where TS is the settlingtime.

Figure 3-6 Settling time [3].

3 - THEORY

16

The Laplace transform of the error (ε in Figure 3-5) becomes:

)s(V)s(A1

1)s( INf+

=ε

If the input signal is a step with amplitude one and the final value theorem isapplied [3], it results in the following expression for the error.

00st

e)0(A1

1

s

1

)s(A1

1slim)t(lim =

+=

+=ε

→∞→ ff

This error (e0) is called the “error coefficient for linear settling”. For thiserror to be zero, the DC-gain of the amplifier (A(0)=A0) should be infinitelylarge.

From the expression above the minimum DC-gain that the amplifier need, tosettle within a certain range can be derived (if linear settling is assumed).

00 e

1AgainDCA(0)large)0(A

f≈==⇒

3 - THEORY

17

3.6.2 Linear SettlingA first order system or a system approximated with a first order systemundergoes linear settling, i.e. the output looks something like in the figurebelow.

Figure 3-7 Output from system undergoing linear settling.

For a system with linear settling the following expression is valid for theoutput [3][9].

−=−

settτ

t

e1K)t(VOUTTdB3

sett11

βω=

ω=τ

−

β is equal to f in Figure 3-5 and settτ is the settling time constant. K is just a

constant and ω-3dB and ωT are the 3dB-bandwidth and unity gain angularfrequency respectively.

From the expression above, the unity gain frequency (fT) can be derived.

t(s)

Mag

nitu

de

3 - THEORY

18

τπβ

=πβτ

=⇒

n

t2

1

2

1f

settsettT

nτ = number of time constants it takes to settle within the [-P,P] range (seeFigure 3-6).

=τ P

1lnn

By assuming that the Operational Transconductance Amplifier (OTA)undergoes linear settling, both fT as well as A0 (DC-gain) can be derived, ifboth the largest acceptable error as well as the range that the output signalshould settle within are known. This will be used during the design phase ofthe OTA.

3.7 Minimum Theoretical Power Consumption for T&H

The most basic track-and-hold circuit only consists of a switch and acapacitor.

Figure 3-8 Basic track-and-hold circuit.

If the switch and the capacitor are assumed to be ideal, i.e. they do notconsume any power due to parasitics and other non-ideal effects, theminimum theoretical power consumed when performing track and hold canbe calculated. The worst case occurs when the maximal input voltage issampled on the capacitor (CS) every period. According to Nyquist’ssampling theorem, the maximum input frequency is half the samplingfrequency (fS/2). If it is also assumed that the maximum input voltage isequal to the supply voltage (Vdd), so that VFS is equal to Vdd/2, and that thecapacitor is completely discharged between every sample, it results in the

following expression for the power consumed2f2

ddVCP sSworstmin, =

Assuming: CS = 1.5pF, Vdd = 1.8V, fS = 80MHz, gives: Pmin,worst = 0.2mW

3 - THEORY

19

3.8 Common Mode Feedback (CMFB)

One of the drawbacks with fully differential amplifiers is that their common-mode output level is not constant. It varies in time and introduces distortioninto the circuit. Common-mode feedback (CMFB) is therefore almost alwaysneeded and used to suppress the common-mode variations at the output, andthereby stabilising the output common mode level.

In the figures below it is clearly shown why common mode feedback shouldbe used. When it is not used the signal is strongly modulated by the commonmode variations, but when CMFB is applied the common mode variation aresuppressed and the modulation is thereby lower. The scales are the same inthe figures below to make it easier to compare and understand why CMFBshould be used. The CMFB used in this design is found in Figure 6-6.

(a) (b)

(c) (d)

Figure 3-9 (a)−−−−Signal before CMFB (b)−−−−Signal after CMFB(c)−−−−Spectrum before CMFB (d)−−−−Spectrum after CMFB

3 - THEORY

20

3.9 Different A/D Structures

This chapter briefly presents four of the most commonly used A/D converterstructures, for example the pipelined A/D converter, in which this T&H aresupposed to be used. More information can be found in [4], for those whoare interested.

3.9.1 Flash ConverterAn N bit flash converter consists of 2N comparators, a resistor string andsome decoding and compensation circuitry. The large number ofcomparators and the fact that the amount grows fast as the number of bits isincreased, makes this solution appropriate for A/D converters up to the sizeof about 8 bits, otherwise the circuit becomes to large. One good thing withthis solution is that it is very fast (one sample per cycle), compared to othersolutions. Another advantage is that a T&H is generally not needed in frontof this type of converter.

The flash converter is often used in the sub stages (residue stages) of apipelined converter, since the resolution of the A/D in each stage is about 2bits the size will not be very large.

Figure 3-10 3 bit Flash A/D converter [4].

3 - THEORY

21

3.9.2 Successive Approximation ConverterThe successive approximation converter has got a moderate circuitcomplexity and yet a reasonably quick conversion time, which makes it oneof the most popular approaches for A/D converters.

Figure 3-11 Successive approximation A/D converter [4].

The converter applies a binary search algorithm to find the correct binaryoutput value. In the first step b1 is found and in the next b2, and so on until allbits are determined.

3.9.3 Algorithmic ConverterThe algorithmic converter requires a small amount of analog circuitrybecause it repeatedly uses the same circuits to perform the conversion. Oneof the drawbacks is that an accurate amplifier with a gain of 2 is needed ifhigh resolution is requested.

Figure 3-12 Algorithmic A/D converter [4].

It works in much the same way as the successive approximation converter,but an algorithmic converter doubles the error voltage instead of changingthe reference voltage. A flow graph of the algorithmic approach can be seenin Figure 3-13.

3 - THEORY

22

Figure 3-13 Flow graph for the algorithmic converter approach [4].

3.9.4 Pipelined ConverterAn N bit pipelined converter consists of N number of residual stages, as theone pictured in Figure 3-14. The idea is to use the concept of the algorithmicapproach, but instead applying pipelining to increase the processing rate toone sample per cycle. Each stage will then look something like the circuitpictured in the figure below.

The complexity will thereby increase compared to the algorithmic converter,but the complexity only grows as N (where N is the number of bits) and notas 2N, which was the case for the flash converter. It is this type of converterthe T&H is designed for.

Figure 3-14 Residual stage of pipelined A/D converter [4].

3 - THEORY

23

3.10 Switch Characteristics of MOSFET

The design of the switches is crucial for the performance of the overall track-and-hold circuit. A poorly designed switch can degrade the performance byfor example introducing harmonics and some other un-ideal behaviour thatwill be presented in the chapters below. Several strategies to avoid orminimize the unwanted effects can be chosen and some of them will bepresented in the chapters that follows next and in chapter 4.3.

3.10.1 Charge InjectionA simple MOS-transistor is often used as a switch. When it is open, thecurrent can flow freely through it and when it is closed the current is turnedoff. The current is equal to the amount of charges per time instant passingthrough the channel, and the amount of charges in the channel isapproximately ( )thgsox VVC − . When the MOS-switch turns off, the charges

in the channel are pushed out of the channel to the drain and source sides.This results in a charge injection error if for example the sampling capacitoris next to the switch. The magnitude of the error-voltage is approximately

( )sampleC2Q , if it is assumed that the channel charge are divided equally

between the drain and source side.

3 - THEORY

24

(a)

(b)

Figure 3-15 MOSFET showing charge injection.(a) – Current through channel before charge injection.(b) – Charges in channel repelled to drain and source (chargeinjection).

The effect of charge injection can be reduced or eliminated by choosing anappropriate circuit structure. Dummy transistors or bottom plate sampling(see chapters below) can for example be used.

3 - THEORY

25

3.10.2 Dummy TransistorsA Dummy transistor is simply a transistor that is shorted between the drainand the source. It is introduced between the MOS switch and the samplingcapacitor. Assume that the charge that is pushed away from the switch oneach side is half of the total charge in the channel of the switch. Then atransistor of half the size of the switch, that is turned on when the switchturns off can store the ejected charges and prevent them from building up anoise-voltage on the sampling capacitor. This is illustrated in Figure 3-16.

Figure 3-16 Basic T&H with dummy transistor.

3.10.3 Bottom Plate Sampling

Figure 3-17 Bottom plate sampling circuit.

The bottom plate sampling circuit above is a simple example on how such acircuit can look like [10]. At the sampling instant, the switch S1 that isconnected to the bottom plate opens before S2, thereby decreasing the errordue to charge injection.

3 - THEORY

26

To better understand the error reduction when using bottom plate samplingone should observe that when the transistor turns off, the charges in thechannel are injected to either the source or the drain side, depending on therelative impedance [11]. Low impedance attracts more charges. Hence, at theopening of S1, most of the charges will be injected to ground and when S2

opens, it will inject most charges back to the input (assuming that the inputhas lower impedance than the output). The effect is that the charge injectionis reduced compared to the case when a simple track-and-hold circuit is used.

3.10.4 Clock Feed Through

Figure 3-18 Simple NMOS-switch with parasitic capacitances.

Parasitic capacitances between gate-drain and gate-source are always presentwhen using simple MOS switches. The clock signal is feed through theseparasitic capacitors and introduces errors. The errors can be suppressed orcancelled by using a differential approach of the track and hold.

3.11 Offset Cancellation

There always exists an offset error between the inputs of an operationalamplifier (OP-amp) due to mismatched transistors in for example the inputstage. Since this is due to process variations, the offset cannot be determinedin advance. Hence, if the offset error is critical for the performance of thecircuit, as for example with high-resolution track-and-holds, compensationfor this error is needed. This can be done by using for example correlateddouble sampling (CDS) [5]. CDS means that the offset is sampled (saved)during one clock phase. This value is subtracted from the offset in the nextclock phase. As a result the offset will ideally be zero. An example of acircuit using correlated double sampling can be found in Figure 3-19.

3 - THEORY

27

Figure 3-19 Example of correlated double sampling (CDS) [5].

Vin is first sampled on C3. In the next phase, a dummy amplification is donewith C3 and C4, generating an estimate of the non-zero virtual groundpotential. At the same time is Vin minus this error-voltage sampled on C1. Inthe last phase, the correct voltage appears at the output since the correctamount of charges is transferred to C2. In other words, the negative error-voltage sampled in the second phase adds to the positive error-voltage duringthe last phase. Hence, the total error is zero.

Another good thing with CDS is that the f1 -noise, which can be a majornoise source in MOS circuits, is reduced [6]. This can be understood bynoting that the offset is ideally a constant value (low frequency signal). Sincethis “signal” is suppressed, then all low frequency signals referring from theOP-amp must be suppressed. Since the f1 -noise is largest for lowfrequencies it will be reduced.

3 - THEORY

28

Figure 3-20 Noise power for a switched-capacitor voltage amplifier [6]:(a)−−−−Noise without offset cancellation.(b)−−−−Noise with offset cancellation.

4 - BRIEF PRESENTATION OF DIFFERENT STRUCTURES

29

4 BRIEF PRESENTATION OF DIFFERENT STRUCTURES

This chapter briefly present different structures to choose among whendeciding what topology to use for the parts in the track-and-hold. Some ofthe advantages and disadvantages with each structure are presented and arelater used in chapter 5 to decide what structures to use in this design.

4.1 Track-and-Hold

In this chapter are some of the most commonly used approaches for track-and-hold circuits presented. This chapter is later on referred back to whendeciding the approach for this design.

4.1.1 Why Fully Differential?Differential structure has several benefits over single ended structures. Thecircuit is less sensitive to common mode noise and the clock-feed-througherror is ideally zero and finally that the even order distortion tones aresignificantly reduced.

Some drawbacks are that the layout gets more complicated since the wiresare doubled and a differential OTA (or OP-amp) need common modefeedback (CMFB) to stabilize the common mode level at the output. Thecommon mode feedback results in a more complex amplifier, which candegrade its performance to some degree, for example the speed.

4.1.2 Open- and Closed-LoopIn an open-loop T&H architecture you do not have any feedback loops. TheT&H consists mainly of some sort of sampling switch, a sampling capacitorand a buffer on the input and output.

Figure 4-1 Track-and-Hold in open loop configuration [12].


30

This simple architecture makes it possible to design for very high speed, butsince it does not give the benefits of feedback, the accuracy cannot be veryhigh.

In a closed-loop T&H architecture on the other hand, a feedback loop isincluded in the circuit for example between the output and input as below.

Figure 4-2 Track-and-Hold in closed loop configuration [12].

The drawback with this is that the speed is reduced, but the accuracy can behigher.

4.1.3 Switched-CapacitorThe signals are represented by the voltages over the capacitors in this kind ofcircuits. The voltages are switched between the capacitors in such way thatthe circuit perform the requested function, for example sampling anincoming signal and deliver the sampled values at the output.

4.1.4 Switched-CurrentIn switched-current circuits, currents instead of voltages represent thesignals. This feature allows the use of only transistors in the circuits and doesnot require capacitors, but even though this technique has been known forsome time, there are very few circuits designed using this technique. Thefocus was therefore on the more commonly used switched capacitortechniques.

4.2 Operational Transconductance Amplifier (OTA)

Five commonly used OTAs are briefly presented in this chapter. They are allfully differential, but the CMFB are not included in the figures. Theperformances of four of the most probable candidates are summed up inTable 4-1 at the end of this chapter.


31

The difference between an OTA and an OP-amp is that the OP-amp has gotan output buffer so that it is able to drive resistive loads. An OTA can onlydrive capacitive loads.

4.2.1 Single-StageThis is the least complex OTA, and hence its speed can be very high. Thedrawback is that the gain is rather low. There will not be possible to reach again of 68 dB, which is required for the amplifier in the T&H (see chapter6.1.1.5).

Figure 4-3 Single-stage OTA.

4.2.2 Two-StageBy adding another stage you get a two-stage amplifier. This modificationincreases the gain as well as the complexity. The increased complexity willreduce the speed a lot, which makes it unsuitable in this design. Thecompensation circuitry (RC, CC) is also included in Figure 4-4.


32

Figure 4-4 Two-stage OTA.

4.2.3 Telescopic CascodeThe reason why the gain of the single-stage OTA is low is that it has got lowoutput impedance. One way of increasing the impedance is to add sometransistors at the output. The transistors are called “cascodes”, and willincrease the output impedance and thereby increase the gain.

Figure 4-5 Telescopic Cascode OTA.

To conclude, the telescopic cascode has got high gain as well as high speed,but by adding more transistors the voltage swing at the output is reduced.This is not desirable, since the supply voltage is low in this particular case.


33

One other drawback is that the common mode voltage at the input cannot beequal to the common mode voltage at the output.

4.2.4 Regulated Cascode (Gain Boosting)Regulated cascode (or gain boosting) can be used to even further increase thegain without decreasing output voltage swing.

Figure 4-6 Regulated cascode (gain boost) OTA.

By applying this method the gain is increased by approximately the gain ofthe gain boost amplifiers. The drawback is that these extra amplifiers mightreduce the speed of the overall amplifier. Hence, they should be designed tohave a large bandwidth.


34

4.2.5 Folded CascodeThe folded cascode amplifier is in a way some sort of a compromise betweenthe two-stage amplifier and the telescopic cascode amplifier. It permits lowsupply voltage, still having a rather high output voltage swing and the inputand output common mode levels can be designed to be equal. Its gain islower than for the two-stage and its speed is lower than for the telescopiccascode, which makes it a good compromise between these two amplifiers.

Figure 4-7 Folded cascode OTA.

The performance of four of the most interesting amplifiers is compared inthe table below.

Gain Outputswing

Speed Powerconsumption

Noise

Telescopiccascode

Medium Medium1 Highest Low Low

Foldedcascode

Medium Medium High Medium Medium

Two-stage High Highest Low Medium Low

Regulatedcascode

High Medium Medium High Medium

Table 4-1 Performance comparison of four different amplifiers.

1 The output-swing of the folded cascode is comparable but somewhat larger, which is desirable.


35

4.3 Switches

Here are the two simplest types of switches presented, MOS switches andtransmission gate switches.

4.3.1 Single MOS SwitchThe maximum output voltage that an NMOS transistor can deliver isapproximately equal to thdd VV − , or equivalently the maximum inputvoltage is approximately equal to thdd VV − . At the same time, the minimum

output (input) voltage is approximately equal to thV (see [9] for more

details).

One can see this as if the resistance (on-resistance of the switch, ONR ) isvarying as a function of the input voltage according to the followingexpression (NMOS transistor):

( )

( )[ ]Ω

−−µ=

⇒==−−µ

=

thINGoxn

NMOS,ON

INS

thSGoxn

NMOS,ON

VVVL

WC

1R

VVVVV

L

WC

1R

From the expression above it is clear that the resistance of a switch are non-linear and that it approaches infinity when VIN approaches Vdd-Vth, which isthe upper limit for the NMOS transistors, as mentioned above. This must betaken into account when designing the switches for the T&H so that thedistortion is minimized.

RON for a PMOS transistor is similar, but instead it approaches infinity forlow voltages (|Vth|). The expression for the on-resistance of the PMOStransistor is found below. Illustrations of RON are seen in Figure 6-9 andFigure 6-10.

( )[ ]Ω

−µ=

thINoxn

PMOS,ONVV

L

WC

1R


36

4.3.2 Transmission Gate SwitchTo circumvent the above problem with a varying switch resistance thebenefit of NMOS for low input voltages and the PMOS for high inputvoltages can be utilized. It is done simply by connecting them in parallel andthereby forming a transmission gate.

Figure 4-8 Transmission gate or complementary switch.

The on-resistance of the circuit above will have the following appearance.

Figure 4-9 Resistance of transmission gate as a function of input voltage.

The drawback of the above structure is that complementary clock signals areneeded, thereby doubling the amount of clock signals. This can betroublesome if the T&H contain many switches.

Mag

nitu

de[ Ω

]

VIN


37

4.3.3 Bootstrapped SwitchAbove we have concluded that the on-resistance of a switch is dependent onthe input signal, and for an NMOS switch its on-resistance is reduced whenthe gate voltage is increased. Finally the instant where the switch opens orcloses is dependent on the voltage difference between gate and input.

It is therefore desirable that the voltage difference between gate and input isas large as possible and of the same magnitude independently of themagnitude of the input signal. This can be accomplished by applying thebootstrapped switch technique.

By storing the clock signal on one phase and add the stored value to theclock signal during the next phase the wanted effect can be accomplished.The drawback is that the gate voltage might be too large and therebyreducing the lifetime of the circuit. The complexity of the switches is alsoincreased, which might affect the performance of the circuit. For exampleincreased power consumption.

In the figure below, an example of a bootstrap switch is shown. Thisparticular circuit does not try to decrease the resistance by increasing thegate voltage. The purpose with this circuit is to always have the same gate tosource voltage (VDD in this case) during the track-phase and thereby getting alinear switch. This is accomplished with the extra buffer connected to CB

during the track-phase.

Figure 4-10 Open loop Track-and-Hold circuit with a bootstrap switch drivercircuit [12].


38

4.4 Clock Generator and Clock Driver

The clock generator has to generate the requested clock signals needed forthe T&H as well as the rest of the A/D converter. In this design it only needto generate the clock signals requested for the T&H. The clock generator canbe found in chapter 5.4.

The purpose of the clock driver is to convert the incoming clock signal (sineshaped), to something close to a square wave, which is the input signal to theclock generator. The load is therefore set by the clock generator, which theclock driver must be able to drive. The clock driver can be found in chapter5.5.

5 - MOTIVATION OF SELECTED STRUCTURE

39

5 MOTIVATION OF SELECTED STRUCTURE

The structure of the T&H as well as the chosen structure of the separatecomponents will be presented and motivated in this chapter. It is based onthe theory and the comparisons of the different circuit solutions presented inthe earlier chapters.

5.1 Track-and-Hold

The error due to input offset of the OTA need to be reduced. The way ofdoing this is to save the offset voltage during the track-phase and thensubtract this during the hold-phase (i.e. correlated double sampling, CDS).This will ideally give offset cancellation. The above technique implies thatone uses a T&H based on switched-capacitor technique.

Another method is to use feedback to reduce the effect of the offset error.The principle is to have a layout that produces an output signal lookingsomewhat like the one below

Where A is equal to the gain. With a large gain the offset error becomessuppressed. This solution can also be accomplished with switched-capacitortechniques.

Based on what have been mentioned above, a switched-capacitor T&H willbe used, the structure found in both [2] and [13] was chosen (see Figure 5-1).The reason is that it is a simple structure with few switches and with a simpleswitching scheme. The drawbacks are that the output voltage varies a lotbetween track and hold phases, since the output of the OTA is set to zero inevery track-phase, and the chosen structure is not suitable for OTAs thatcannot be shorted between input and output, which is the case when usingfor example telescopic cascode.

A

VVV offset

inout ±≈


40

Figure 5-1 Selected T&H structure.

5.2 OTA

In Table 4-1 (page 34) it shows that the two-stage amplifier probably will beto slow for this application and will therefore not be used. What is left is thento choose between the telescopic cascode and folded cascode. In a later stagewhen the chosen amplifier has been simulated and its gain has beenmeasured, the decision whether to use regulated cascode or not will be taken.

In the same table it is quite clear that the telescopic cascode is a bettersolution than a folded cascode if speed, noise and power consumption aretaken into consideration. The drawback with the telescopic cascode is thatthe voltage swings are slightly lower and it does not permit short-circuitbetween input and output, which is needed due to the choice of T&Hstructure.

One additional drawback with telescopic cascode is that it might need anadditional stage on the output to increase the voltage swing. Due to this, onemore CMFB is needed, which consumes more power, and the output stagealso adds to the power consumption. At the same time the second stage willreduce the speed of the overall amplifier.

Hence, a folded cascode topology of the amplifier was chosen and tominimize the power consumption the simple CMFB found in [6] were used.The chosen amplifier is seen in Figure 5-2.


41

Figure 5-2 Folded cascode amplifier.

The two extra transistors (M1CMFB, M2CMFB) reduce the allowable outputsignal swing. Fortunately, as later mentioned in chapter 6, this can beaffordable since the common mode level is chosen relatively large (1.05Volt).

5.2.1 Why NMOS or PMOS at Input-Stage?When deciding what kind of transistors to use as input-transistors, severalthings have to be taken into account.

i. The mobility of the electrons is about three times larger than the mobilityof the holes. gm, and thereby the gain and unity gain frequency will behigher when using NMOS- instead of PMOS-transistors, for the sametransistor sizes.

ii. When using NMOS transistors the impedance at the folding points (A, Bin Figure 5-2) will be lower, due to the lower impedance of the PMOStransistors at the folding points. Both the gain as well as the phasemargin will be lower than if using PMOS transistors at the input, for thesame sizes of the transistors.

iii. NMOS transistors have lower thermal noise.


42

iv. PMOS transistors have lower f1 -noise.

Since high gain is needed, NMOS transistors at the input were preferred overPMOS. Also, since correlated double sampling will be used, the f1 -noise isreduced, which also reduces the need of using PMOS transistors instead ofNMOS. It might anyway be worth designing another OTA with PMOS at theinput, to see which one that seems to be the best solution in this particularcase.

5.2.2 Bias Circuit for the OTA

Figure 5-3 Chosen structure for bias circuitry.

The benefits of this circuit are that it has got constant transconductance dueto Rb and at the same time the bias loop consists of wide-swing currentmirrors. By using the wide-swing current mirrors the output resistance isincreased, which greatly reduces the harmful second-order imperfectionscaused by the finite-output impedance of the single transistors and at thesame time not restricting the signal swing too much. More details can befound in [4].


43

The bias loop sets the bias current in the circuit and V1 and V4 are also set bythis part of the bias circuit. The size of M7b is designed to be four timeslarger than M8b, the current through this loop will then be amplified. To stopthe current from increasing and stabilize it to the desired value, Rb is added.As the current increases, the voltage drop over Rb will increase and at thesame time will Vgs of M7b decrease. The current in this circuit will therebyhave a certain maximum value and since the gain is positive, this value willstay stable. The bias loop is connected together with the cascode bias. Thelatter sets the value of V2 and V3.

The maximum allowable current mentioned above is only one out of twosolutions. The other is that all currents are zero. To avoid this latter solution,the start-up circuitry is used. If the currents are zero, V3 will be zero. Theinverter consisting of M31b and M32b then turn on M33b and M34b. They will inturn inject current into the bias loop, which will start up the whole circuit.Then V3 starts to increase and the inverter will turn off the start-up circuitry.

5.3 Switch

There are several different switches to choose among. The demands are thatit should not consume any or very little power and at the same time notdegrade the performance too much. It is also desirable that the complexity isminimized. This is why a simple NMOS switch was chosen as a firstapproach and if needed, apply the technique of transmission gates later in thedesign-flow.

5.4 Clock Generator

The clock generator structure is set by the demands on the clock signals, andsince a simple structure of the T&H is chosen, the clock generator onlyconsists of three inverters [9]. If transmission gates will be used for switches,an inverted version of each clock-signal will be needed.

Figure 5-4 Clock generator.


44

It is also important to minimize the delay between the individual clocksignals, otherwise the settling time of the output of the T&H will be largerand the exact sampling instant will be more uncertain. See the figure belowwhere the delay between Φ1 and Φ3 is 600 ps and 100 ps respectively. TheSFDR is also about 3 dB lower when the delay is 600 ps compared to whenthe delay is 100 ps. The T&H structure used is found in Figure 5-1 and Vout

in the figure below is equal to Vout+-Vout-.

Figure 5-5 Comparison of output signal of T&H (and its frequency spectrum)when delay is 100ps (a,c) respectively 600ps (b,d).

(a) (b)

(c) (d)


45

5.5 Clock Driver

The demand on the clock driver is that it must be able to drive the clockgenerator. It should also generate a single ended output signal, which is closeto a square wave, from the sine shaped differential input signal.

The demands on this circuit are quite loose. A simple single-stage single-ended amplifier with differential input is used in this design. To furtherincrease the drive capabilities and shorten the rise and fall times, the outputfrom the clock driver is buffered with a couple of inverters with increasingsize for each step (as illustrated in the figure below). The whole clock driver,including inverters, is found in the figure below.

Figure 5-6 Clock driver with inverters as buffers.

6 - DESIGN

46

6 DESIGN

This chapter describes the design phase, how the specifications on thedifferent building blocks were found and how they were designed andverified. The design phase were roughly divided into the following steps

1. Finding specification of the OTA.

2. Simulation and performance verification.

a. If the gain is to low the gain boost amplifiers must be designed andadded to the amplifier.

b. Simulation and performance verification of amplifier with gainboost amplifier.

3. Design of common mode feedback (CMFB).

4. Verification of the performance for the whole amplifier.

5. Simulate the T&H with the amplifier and verify whether pass transistorgates are needed or if the switches can be made of single NMOStransistors.

6. Design of the switches.

7. Design of the bias circuit and verification of the performance of theamplifier using the bias circuit.

8. Design of the clock generator. Finding its specification is also includedin this step.

9. Finally the design of the clock driver.

The T&H was then simulated and its performance is presented in chapter 7.

6 - DESIGN

47

6.1 Operational Transconductance Amplifier (OTA)

This chapter describes how the specifications for the OTA were found andhow it was designed. The design of CMFB and gain boost amplifiers canalso be found in this chapter.

Figure 6-1 Folded cascode OTA including gain boost amplifiers and commonmode feedback.

6.1.1 Finding Specification for the AmplifierThe specifications on the amplifier were not given in advance, so they arederived in this chapter. More details and theory can be found in for example[14].

6.1.1.1 Signal Levels and Sampling CapacitorSince this T&H is supposed to be used as a front-end of a A/D converter, the

size of the sampling capacitor should be chosen in order to reduce itsC

Tk B -

noise [15], the signal magnitude must be known to be able to do this.

6 - DESIGN

48

If the signal magnitude is large, the sampling capacitor can be smaller,thereby reducing the load on the OTA, which then can be designed for alower bias current and hence consume less power. On the other hand, sincethe supply voltage is rather low (1.8V) and since a large signal swing mightresult in increased distortion in the switches, a signal swing of about 600 mV(FS) seemed like a good compromise. The value of the sampling capacitorwas then calculated (see below)

mV18922

VV

mV53510600dBFS1atVandV

mV600FSV

min,pp,OUTrms,OUT

20

1

pp,OUTpp,IN

OUT,IN

==⇒

⇒=×≥−

⇒=−

Now when the output rms-value (root mean square) of the signal is known,the value of the sampling capacitor can be calculated (the noise voltage inthe denominator are derived in Appendix I)

pF2.1C

C

Tk226.1

Vlog10dB66 S

S

B

2rms,OUT10 ≥⇒

×××=

The sampling capacitor was chosen to be 1.5pF to get some margin for theSNR. Theoretically the SNR should be 67dB instead of 66dB with thischoice of capacitor.

According to the results in Appendix I, theC

Tk B -noise is the dominating

noise source and thereby defines the noise floor.

(rms)V74C

Tk2

S

B µ=

The “output DC-offset” must be smaller than 74µV, about 50µV wereselected to get some margin. To conclude

µV50offsetDCOutput

mV600VV

pF1.5C

FSOUT,FSIN,

S

=−

===

6 - DESIGN

49

6.1.1.2 Slew Rate (SR)To find the required slew rate, the time allocated for slewing during eachperiod must be known. The rule of thumb from [14] was used to find theslewing time. It states that the time allocated for slewing should be about ¼of half the sampling period (TS/8), which is 1.5625ns when fS = 80MS/s.

The outputs of the OTA should be able to deliver a 0.3 Vpp signal (half thefull scale signal) and this is also the highest voltage step allowed. Hence,slew rate can be calculated as follows:

µsV192800.38f8V

T

V8SR Spp

S

pp =××===

6.1.1.3 Load Capacitance (CL,tot)In this chapter the total load capacitance is calculated. This is the load, whichthe amplifier must be able to drive.

Figure 6-2 Track-and-Hold with parasitic capacitances included.

The largest capacitance seen from the output of the amplifier occurs duringthe track phase, when Φ1 and Φ3 are closed and Φ2 is open. The capacitanceloading the amplifier becomes

SGSLtot,L CCCC ++=

6 - DESIGN

50

CGS was assumed to be at least less than 0.4pF, which was confirmed by thesimulations, and CS was chosen to 1.5pF (see above). CL was the loadcapacitance and was assumed to be equal to the sum of CGS and CS (worstcase). This is reasonable, since the circuit was supposed to be used in apipelined A/D-converter and each stage of that converter has a T&H on theinput (see Figure 3-14). If some other solution was used the capacitance willbe lower. CL,tot then becomes

pF3.8C totL, =

6.1.1.4 Unity Gain Frequency (fT)

Since 8TS was allocated for slewing, only 8

T3 S was left for settling. Linear

settling was assumed and the results from chapter 3.6.2 were used, fT wasthen calculated according to below. To get the minimum required fT, thelowest β was used for the calculation. holdβ and trackβ can be found fromexamination of Figure 6-2. The requested accuracy (P) was calculated byassuming that the signal should lie within 1LSB after the settling time. This

is equal to P= 111021 22 −− = .

1track ≈β

( )MHz3244.05f

0.750.82π2ln2f

f S

11S

T ==××

>⇒

8.0CC

C

PS

Shold ≈

+≈β

6.1.1.5 DC-Gain (A0)The following expression for the linear settling error coefficient was derivedin chapter 3.6.1 (f=β).

00 A1

1e

β+=

This error should be less than ½LSB, giving

6 - DESIGN

51

dB68.2β

2A

22A

1e

11

0

111021

00

≈>

⇓

=<β

≈ −−

In general this becomes:

( )Bits)of(NumberNOBNwhere,

2A

1N

0 =β

>+

6.1.1.6 Phase MarginSwitched capacitor circuits are in general designed for a phase margin ofabout 60 to 75 degrees [4]. In this design it was assumed that it undergoeslinear settling, i.e. a phase margin of 90 degrees, when the specifications ofthe amplifier was derived. This was why the phase margin should not bechosen too low. In the figure below, the output of the T&H is shown for twodifferent phase margins. As expected, the settling time as well as the ripple isdecreasing as the phase margin is increasing. The reason is that the systemacts more and more like a system undergoing linear settling. A phase marginof about 65 degrees were first chosen, but after some simulations of the T&Hit was clear that the phase margin should be around 70 degrees to getacceptable settling times. This also increased the SFDR a few dB.

(a) (b)

Figure 6-3 Settling time for different phase margins(a) −−−− 67 degrees, (b) −−−− 76 degrees.

6 - DESIGN

52

6.1.2 Calculation of Lower Limit ValuesSome approximate values were calculated prior to the simulations to getsome feeling for the different sizes and to know the lower limit to someimportant parameters.

6.1.2.1 Bias CurrentsSince the slew rate (SR) is approximately equal to the available bias currentover the total output capacitance (CL,tot), a lower limit for the bias currentscould easily be calculated.

µA800103.810192CSRII 126totL,b2b1 ≈×××=×=+ −

As a first approach the following currents were chosen:

A640I4I 2b1b µ==

After some simulations it was clear that the current had to be increased tofulfil the demand on the phase margin. Thus, the following currents was usedin the end

A350IandA790I 2b1b µ=µ=

6.1.2.2 gm of the Input TransistorsThe unity gain frequency is approximately equal to the product of the DC-gain and the lowest pole. This is used below to calculate the minimumneeded gm of the input transistors.

mS7.3g

103242C

g

C

g

g

gpA

1m0,

6

tot,L

1,0m

L

out

out

1,0m10u

>⇒

⇒××π>==×≈ω

6.1.2.3 Minimum Transistor SizesFinally the minimum transistor sizes to accomplish high enough gm0,1 andVds,sat were calculated. This can be done with help of the followingequations.

6 - DESIGN

53

1,0d1,0

21,0m

1,0 IK2

gS =

2i,sat,dsi

dii

VK

I2S =

Wherei

ii L

WS =

The resulting sizes (Si) are as follow.

S0,1 ~ 151 S10,11 ~ 2.04S2,3 ~ 243 S12,13 ~ 11.4S4,5 ~ 243 S14,15 ~ 50.2S6,7 ~ 108.4 S16,17 ~ 27.1

Table 6-1 Minimum transistor sizes.

When using the above sizes (L=0.27 µm) the result was not at all satisfying.

Figure 6-4 Bode diagram of amplifier using the above values.

6 - DESIGN

54

This approach, just trusting the theoretical values, is not likely to succeed.Neither will an approach just trusting on simulations work, since the deepunderstanding of how the circuit works and which parameters that affectwhat property of the amplifier, will be lost. A combination of these twoapproaches must be used.

Before starting to simulate it must be known what is to be accomplished andhow it should be done. The demands on unity gain frequency, phase marginand slew rate are all known. The latter two are mainly depending on thechoice of bias current. The approach for the simulations was therefore to firstchoose a bias current that theoretically satisfied the slew rate and then verifyby simulations that the phase margin was within a reasonable range from thedesired value. Then it was time to adjust the parameters to maximize thegain, or at least get a gain that matched the requirements. How this was doneis explained in the chapters that follow.

To get some lead to in what direction the different parameters should bechanged, to enhance the performance, the following table showing thedependencies was derived. The derivation can be found in Appendix II.

A0 fT p1 p2

WM0,1 ↑ 1↑ 1↑LM0,1 ↓ 1↑ 1↑ 2↑IdM0,1 ↑ 1↓ 1↑ 2↑IdM10,11 ↓ 3↑ 3↓ 1↓WM10,11 ↓ 1↓ 1↑LM10,11 ↓ 1↓ 1↑LM12,13 ↓ 2↓ 2↑LM16,17 ↓ 2↓ 2↑WM14,15 ↓ 1↓ 1↑ 1↑LM14,15 ↓ 1↓ 1↑ 3↑CLoad ↓ 2↑ 2↑

Table 6-2 Table showing how different parameters affect the DC-gain (A0), unity gainfrequency (fT), 3dB bandwidth (p1) and the second lowest pole (p2). Highernumber next to the arrows means stronger dependency.

6 - DESIGN

55

6.1.3 Vgs of Input TransistorsThe first observation was that both A0 as well as the unity gain frequencyincreased when gm0,1 increased. Hence, gm0,1 should be maximized, but stillmaintaining the desired bias current. Maximizing gm0,1 when the current wasconstant was equal to maximizing the size of the input transistors, whichwere done by choosing Vgs as low as possible. This can be understood byexamination of the equation below. It is clear that when decreasing Vgs thesize must be increased to maintain the same current. The size is included in β( L

W∝β ).

( ) ( )ds2

thgsd V1VV2

I λ+−β=

6.1.4 Input and Output DC LevelsDue to the choice of T&H structure the OTA needed to be designed so thatits input and output DC levels are equal, since the output and input of theOTA are shorted together during the track phase.

The lowest DC value was set by the lowest possible Vgs of the inputtransistors, the required voltage swing and the fact that the voltage drop overM2 and M4 should be as large as possible to get as high output impedance ofthe current mirror as possible. The voltage drop over M2 and M4 should atleast be larger than Vds,sat for each of them, so that they work in the saturationregion.

The largest DC value was set by the fact that the voltage drop over M14 andM16 should be as large as possible to maximize the output impedance andthereby the gain, and of course the voltage swing on the input had to beconsidered.

This implies that some compromising had to be done and in combinationwith some simulations a DC level of 1.05 Volt seemed like a fairly goodcompromise.

6.1.5 Current Mirror Transistor SizesTo maximize the overall gain, the output impedance should be maximized.Maximizing the output impedance can be accomplished by maximizing theoutput impedance of the separate current sources in the output part, whichwere realized by current mirrors.

6 - DESIGN

56

The conclusion from the current mirror chapter was that to maximize theoutput impedance of the current mirror, the lengths should be maximized aswell as W2, the width of the transistor closest to the output of the currentmirror (top transistor in NMOS-mirror and bottom transistor in PMOS-mirror). On the other hand, by using larger lengths at for example the output,the capacitance at the output increases, which reduces the first pole andresults in a poorer phase margin as well as lower unity gain frequency, whichmust be taken into account.

Due to the problem with increased parasitics, the lengths were not increased,but instead W2 was increased. This was accomplished by allocating the Vds-voltages over M2 and M4 (see Figure 3-4) in such a way that M2 was forcedto be larger than M4 to get the correct bias point. M2 and M4 were cascadedand therefore, if W2 was increased then Vds4 would increase. So by choosingVds2 smaller than Vds4, W2 had to be larger than W4 and thereby the outputimpedance was increased, which resulted in higher gain.

To conclude, simply by allocating the available voltage drop over the outputof the current mirror, in a more clever way, the output impedance could beincreased. Vds2 were chosen to be equal or (preferable) smaller than Vds4.

In general the largest possible Vds4 that still allows the rest of the transistorsin the circuit to work in the right region, and that not results in transistorsizes that are unreasonable, should be used. By using this concept and somesimulations, the following Vds voltages seemed reasonable.

Vds0,1 = 960 mV Vds10,11 = 400 mVVds2,3 = 150 mV Vds12,13 = 650 mVVds4,5 = 250 mV Vds14,15 = 310 mVVds6,7 = 440 mV Vds16,17 = 440 mV

Table 6-3 The chosen Vds-voltages for first attempt.

6.1.6 Approach for SimulationsSince the circuit is rather large and has got a lot of parameters, it is notpossible to simulate and vary the whole circuit and parameters at once. Somekind of “divide-and-conquer” technique must be used. In this case this meansthat each transistor should be simulated individually to decide its parametersso that it work in the right region and at the right bias point. When they allwork according to the requirement on bias point, they can gradually beconnected together, still simulating for each step to verify that they still workas they are supposed to.

6 - DESIGN

57

By iterating like this, the final circuit is guaranteed to work at the correctbias point, left is just to check if the performance match the requirements. Ifit is not, the circuit has to be modified by for example changing the biaspoints, which can be done by choosing another bias current or alter the Vds-voltages. Resistors simulated the rest of the circuit to get the correct voltagedrop over the transistors that was to be simulated.

The simulations can basically be divided into the following steps:

1. Decide a bias current.

2. Allocate the Vds-voltages.

3. Introduce resistors of sizes that give the desired Vds-voltage for thedesired bias current.

4. Sweep the parameters to find the values that give the right bias point.

5. Connect the transistors together gradually and repeat step 3 to 5 untilthe whole circuit is connected together.

6. Check that the bias points are correct and verify if the performance ismatching the demands.

7. Repeat from step 1 if the performance is not fulfilled.

By applying the steps above an amplifier that fulfilled the demands on phasemargin and unity gain frequency was reached. The upper limit for the gainseemed to be around 60 dB. Gain boosting (regulated cascode) had to beused to increase the gain further. The design of the gain boost amplifier isdescribed in chapter 6.1.7.

6.1.7 Design of Gain Boost AmplifiersThe overall gain is in general increased by approximately the gain of thegain-boost amplifiers, the gain specifications of the gain-boost amplifierswere thereby known. The unity gain frequency of the gain-boost amplifiersshould be large enough so that they do not affect the frequency behaviour ofthe overall amplifier, at least not too much. They will reduce the unity gainfrequency of the overall amplifier since by adding the gain-boost amplifiersto the output side, extra capacitance and thereby some extra poles are added.The bias current should be as low as possible, still allowing high enoughunity gain frequency. Finally the amplifiers output DC-level was known,

6 - DESIGN

58

since it must be the same as the voltage at the node onto which the outputwas to be connected.

Since the gain only had to be increased by about 12 dB, an inverter waschosen as gain-boost amplifier. It should give high enough gain and it is notvery complex, it should thereby be easy to get sufficiently high unity gainfrequency and it should not degrade the performance of the overall amplifiertoo much.

Figure 6-5 Gain boost amplifiers.

The simulation and sweeping of parameters were performed as in the chapterabove. When they were working according to the demands, they wereconnected to the main amplifier and the performance of the overall amplifierwas verified.

The gain had increased, but both the unity gain frequency as well as thephase margin was degraded too much. The bias currents of the mainamplifier were then increased and some of the bias voltages were altered.Since the output DC-level of the gain boost amplifiers can be chosen from awider range than the bias voltages, this was another benefit of using gainboosting. In a way they give both increased gain and voltage level shift. Byapplying this, Vb3 was increased and Vb2 decreased, which allowed furtheroptimisation of the output impedances.

After a couple of iterations, an amplifier that fulfilled all the demands on theperformance was found. The design of the common mode feedback was leftto do, which is described below.

6 - DESIGN

59

6.1.8 Design of Common Mode Feedback (CMFB)The CMFB architecture in the figure below was chosen, much due to itssimplicity compared to other solutions.

Figure 6-6 OTA with CMFB indicated.

M1CMFB and M2CMFB were working in the linear region and were therebysimply voltage-controlled resistors. Their gates were connected to the outputnodes and when for example the common mode voltage of the output wasincreasing the Vgs of M1CMFB and M2CMFB increased, their resistance therebydecreased. The voltage drop over them then also decreased, which resulted ina decrease of the common mode level on the output, which was desired.

According to the specifications in Table 2-1 the SNR should be at least 66dB. The two CMFB-transistors were then designed to decrease the commonmode variations to a level well below this. The common mode variationsshould then neither affect the SNDR of the amplifier nor the track-and-hold.

According to the simulations, a Vds of 100 mV was enough. Due to this,M12,13 was redesigned to have a voltage drop of 550 mV instead of 650 mV.

This reduced the output impedance of the NMOS part on the output side ofthe amplifier, but it was still much larger than the PMOS part. The gain wastherefore not significantly decreased.

6 - DESIGN

60

6.2 Bias Circuitry

Until now all the bias voltages had been set by voltage sources during thesimulations. The final step was therefore to design the bias circuitry. Thechosen circuit is shown in Figure 6-7 below.

Figure 6-7 Chosen structure for bias circuitry.

6.2.1 Calculation of Approximate ValuesThe voltages that the bias circuit will deliver are known from the simulationof the amplifier. These node voltages were thereby fixed and the differentVds voltages could be derived, only considering that all transistors shouldwork in the saturated region. The magnitude of the bias current in the biascircuit was also decided at this point.

6.2.2 Design and Simulation FlowThe simulation methodology (“divide-and-conquer”) was the same as inchapter 6.1.6. When the circuit delivered the correct voltages it wasconnected together with the amplifier. The simulations of the wholeamplifier were then performed and it was soon obvious that the parasiticcoupling between the bias circuit and the drain of M2,3 (Cgd) was too large.This had the effect that a common mode signal was amplified, thereby givinga poor common mode rejection ratio (CMRR). This problem was solved by

6 - DESIGN

61

decreasing the size of M2,3, which in turn was accomplished by increasingthe allocated Vds voltage, thereby forcing the size of the transistors todecrease. The final Vds voltages are shown in the table below, where also theVds voltage of the CMFB-transistors has been included.

Vds0,1 = 960 mV Vds10,11 = 400 mVVds2,3 = 200 mV Vds12,13 = 550 mVVds4,5 = 200 mV Vds14,15 = 310 mVVds6,7 = 440 mV Vds16,17 = 440 mV

VdsCMFB = 100 mV

Table 6-4 The final Vds-voltages.

Finally, the bode diagram of the amplifier using the values above is shown inthe figure below.

Figure 6-8 Bode diagram of the final amplifier.

6 - DESIGN

62

6.3 Switch Arrangement

The approach for the design of the switches was to first use NMOStransistors. This was not a very good solution. The SFDR became very poor,almost 17dB below the required value of 64dB. The reason for their poorperformance is illustrated in the figure below. The resistance magnitude inthe figures below is not equal to the magnitudes in the simulated switches;these figures only serve as illustrations to explain the ideas.

Figure 6-9 Resistance of an NMOS-switch as a function of the input voltage.

The signal range is between 0.9 and 1.2 Volt. The resistance is relativelynon-linear in this range, as can be seen in the figure above. This is why theNMOS transistor works poorly for high voltages. The demand on SFDR wasnot reached as a result of that.

A PMOS transistor on the other hand, is known to work poorly for lowvoltages and rather well for high voltages. The NMOS-switches wastherefore replaced with PMOS-switches. They had to be about six timeslarger than the NMOS-switches, but the SFDR increased to 52.2 dB, still faraway from the goal. From Figure 6-10 it is clear that the resistance is stillrelatively non-linear.

6 - DESIGN

63

Figure 6-10 Resistance of a PMOS-switch as a function of the input voltage.

As was mentioned in the theory-part, the transmission-gate-switch (NMOS-and PMOS-transistor connected in parallel) might be the solution to theproblems above. The resistance for the NMOS- and PMOS-switch above areused to get the resistance for the transmission gate switch in the figurebelow.

Figure 6-11 Resistance of NMOS-, PMOS- and transmission-gate switch.

As seen, the resistance for the transmission-gate-switch is much more linear.In reality the total resistance will look something like in Figure 4-9 on page36, but Figure 6-11 still illustrates why transmission-gate-switches can be awise choice. The problem was still to design the two transistors so that theswitch was linear enough for this design.

Re

sist

ance

mag

nitu

de

6 - DESIGN

64

Since the circuit is differential, it is the third order distortion that isdegrading the SFDR. If a non-linear resistance according to below isassumed

2

2

22

210dI

RdRIRIRRR ∝⇒++=

Where “I” is the current through the resistor. Assume now that a sine-shapedcurrent is applied to the resistance. The voltage over the resistance thenbecomes

( ) ( ) ( ) ( )tsinRtsinRtsinRtsinIV 32

210R ω+ω+ω=ω==

The third order distortion is due to the ( )tsinR 32 ω -term. If R2 were equal to

zero, the distortion would be zero in this case. This is equal to saying that thesecond derivative of the resistance should be zero. Inspection of theresistance as well as the second derivative of the resistance of the switch wasused during the simulations to compare the different switches of varioussizes.

First the relative size (X) between the PMOS- and NMOS-transistor, i.e. howmany times larger the PMOS-transistor should be, was swept. The curves inthe figure below show that for values of around six, the resistance start tolook fairly linear. The PMOS-transistor was first chosen to be seven timeslarger than the NMOS-transistor.

6 - DESIGN

65

Figure 6-12 Switch resistance and its 2nd derivative as a function of input voltage fordifferent relative sizes (X).

When the relative size was fixed it was time to sweep the size ( LWS = ) of

the NMOS-transistor. The size of the PMOS-transistor was sweep at thesame time, but X times larger (7 in this case). The resistance and its 2nd

derivative are shown in the figure below for some different sizes (S). Byinspection of this figure, the size was chosen to 100.

Figure 6-13 Switch resistance and its 2nd derivative as a function of input voltagefor different sizes (S=W/L).

The T&H was simulated with this switch and it became obvious that theNMOS-transistor had to be somewhat larger. The PMOS-transistor could

6 - DESIGN

66

instead be somewhat smaller. The final switch sizes are shown in the tablebelow.

NMOS PMOSWidth 19.8 µm 118.8 µmLength 0.18 µm 0.18 µm

Table 6-5 The sizes of the transistors in the final switch.

The spectrum of the output signal is displayed in Figure 6-14 to illustrate thedifference in linearity of the T&H when using transmission-gate-switches(using the final sizes) compared to NMOS-switches (W/L=19.8/0.18µm).SFDR is marked in the figure. The SFDR was increased by nearly 17 dBwhen transmission-gate-switches were used, which clearly shows theadvantage of using this kind of switches. The drawback was though that theamount of clock-signals were doubled, thereby increasing the powerconsumption of the clock-generator. The dominating harmonic seen inFigure 6-14 is the third order harmonic folded into the signal band.

(a) (b)

Figure 6-14 Spectrum of output signal from T&H when using,(a) – NMOS-switches, (b) – transmission-gate-switches.

6 - DESIGN

67

Finally is the performance of the T&H when using NMOS-, PMOS- andtransmission-gate-switches respectively compared in the table below. Thisalso shows the advantage of using the latter switch type.

NMOS PMOS Transmission-gateWNMOS 18 µm * 18 µmWPMOS * 108 µm 108 µmL 0.18 µm 0.18 µm 0.18 µmfIN 30 MHz 30 MHz 30 MHzfS 80 MHz 80 MHz 80 MHzSNR 74.3 dB 75.3 dB 75.3 dBSFDR 47.5 dB 47.6 dB 64.3 dBSNDR 47.1 dB 47.2 dB 62.1 dBHold-mode feedthrough -100 dB -99 dB -96 dBPower 7.5 mW 7.6 mW 7.6 mW

Table 6-6 Comparison of performance of T&H for different switches (signallevels at -1dB FS).

6.4 On-Chip Clock Driver and Clock Generator

The clock driver consists of an amplifier with differential input and singleended output, which is buffered with inverters to be able to drive the clockgenerator (Figure 5-6 on page 45). The amplitude of the output from the

amplifier is equal to 2VDD . The output signal was assumed to be sine-

shaped. The maximum required slew-rate of the amplifier could then becalculated, which was used when the required bias current was derived. Thecalculations are shown below.

( ) ( )

Af2AslopeMaximum

tsinAdt

dysignaloutputofSlopetAsinysignalOutput

π=ω=

⇒ωω==⇒ω==

The slew-rate should be equal or larger than the maximum slope; this resultsin the following slew-rate.

6 - DESIGN

68

sV470SRmarginsomegetTo

sV453

MHz80ff

mV9002

VA

Af2SR

max,S

DD

µ=⇒

⇒µ===

===π≥

The buffer stage was simulated to get the parasitic capacitance that is loadingthe clock driver amplifier. It was approximately 125fF. This value is usedbelow to get the bias current.

A59sV470pF125SRCI

C

ISR LOADBIAS

LOAD

BIAS µ=µ×=×=⇒=

This value was nearly doubled to get some further margin, i.e. a bias currentof 100µA was chosen. The output signal of the clock driver is shown inFigure 6-15.

Figure 6-15 Output signal from clock driver.

6 - DESIGN

69

The last stage was to design the clock-generator. The clock-generator firstused is shown in Figure 6-16 and can also be found in [9]. Sincetransmission-gate-switches were needed, the clock signals in the figurebelow had to be inverted to get the clock signals for the PMOS transistors.Since Φ2 is the inverse of Φ1 the circuit could be simplified and is shown inFigure 6-17 together with the clock driver.

Figure 6-16 Clock generator.

Figure 6-17 Clock driver and the final clock generator.

The size of the inverters sets the rise- and fall-times, but if too large invertersare used, the number of buffer stages in the clock driver must be increased,which increases the power consumption.

6 - DESIGN

70

Figure 6-18 Rise- and fall-times for the two cases (delay = 600ps and 100psrespectively). Rise-time about 55% of the delay.

The track-and-hold circuit was simulated for different sizes of the invertersand the result seen in Figure 5-5 was found. The rise time was about 55% ofthe delay time in each case, as shown in Figure 6-18 above. A delay of about100ps was chosen. The clock generator was therefore designed to have a risetime of about 55ps.

The last demand was that the rise- and fall-times should be approximatelythe same. This was accomplished by making the PMOS transistor about 3.06times larger than the NMOS transistor in each inverter.

Figure 6-19 Schematic of the inverters.

6 - DESIGN

71

The resulting sizes are shown in the table below.

NMOS PMOSWidth 25.1µm 76.7µmLength 0.18µm 0.18µm

Table 6-7 The sizes of the inverters in the clock generator.

7 - RESULTS

72

7 RESULTS

The performance of the complete track-and-hold is presented in Table 7-1below. The performance is based on circuit level SPICE simulations inCadence with foundry provided BSIM3 transistor models. The simulationswere done for the worst-case input frequency (30 MHz).

Requirement PerformanceSampling Frequency Range 10 to 80MS/s 10 to 80MS/sInput Signal Frequency Range DC to 30MHz DC to 30MHzHold-mode feedthrough < -65dB -96dBSNR 66dB 75.3dBSFDR 64dB 64.3dBSNDR 61dB 62.1dBTwo-tone IMD 58dB 70.3dBCLOAD * 2.3pF

PT&H1 Minimized 7.6mW

Power ConsumptionPOTA

2 Minimized 6.6mW

Table 7-1 Comparison of the requirements and the simulated performance. The inputfrequency is 30 MHz.

SNR was calculated by dividing the root mean square value of the outputsignal by the integrated noise. The noise was integrated over the range DC to2fin,max (0-60MHz).

The power consumption was found by simulating the T&H circuit with30MHz input frequency and then calculate the average of the current fromthe supply.

The SFDR is found in Figure 6-14b and the two-tone IMD is found in Figure7-1, showing the spectrum for the output of the track-and-hold.

1 The total power of the whole track-and-hold circuit, including the power consumption of the OTA.

2 The power consumption of the OTA.

7 - RESULTS

73

(b)

Figure 7-1 Spectrum where two-tone IMD are found.

The SNDR was calculated by combining the SFDR and the SNR.

The load capacitance was increased from 1.8pF to 2.3pF during the design ofthe amplifier, to get high enough phase margin.

8 - CONCLUSIONS

74

8 CONCLUSIONS

The purpose of this work was to design a low-power track-and-hold circuitwith 10bit resolution, sampling frequency of 80MHz and a 30MHz analogbandwidth using a 0.18µm process with a supply voltage of 1.8Volt. Therequirements that were given are all fulfilled, as can be seen in Table 7-1 onpage 72.

The SFDR as well as the SNDR was the most difficult parameters to fulfil,but after some redesigning of the switches the goals was reached with a0.3dB and a 1.2dB margin respectively. Increasing the size of the switchescan further increase the SFDR. The power consumption will then alsoincrease, mainly due to increased capacitance of the switches.

The other parameters are all fulfilled with a much larger margin. The SNR,two-tone IMD as well as the hold-mode feedthrough are fulfilled with amargin of about 9dB, 12dB and 31dB respectively. These results arehowever expected to be degraded if this circuit is manufactured andmeasured upon, which only makes it desirable to have some margin.

Power consumption may be further reduced, primarily by choosing someother approach. Some observations are presented below that was discoveredduring the design.

8.1 Discussion

At the end it became obvious that more simulations of ideal componentsshould have been done, in for example Matlab. This would have been helpfulto find some of the demands, but since the time was limited, the decision wasnot to look too deep into the problems. It would although be interesting tolook more deeply into the details and try to find ways to optimise the design,if it is possible. That is though out of the scope for this work.

The power consumption is strongly dependent on the choice of samplingcapacitor. If the capacitance is reduced the load capacitance of the OTA willdecrease. The current needed to fulfil the demand on slew rate would thenalso decrease, thereby being able to decrease the sizes of the transistors inthe OTA. A reduction of bias current and transistor sizes will increase thegain and bandwidth of the amplifier, the phase margin would probably alsoincrease.

8 - CONCLUSIONS

75

The drawback of decreasing the sampling capacitor is that the signal swinghas to be increased to maintain the SNR. This would result in more distortionin the switches. Some form of bootstrapping might then have to be used forthe switches. An increased signal swing would also be troublesome due tothe low supply voltage. If the swing is increased too much, a rail-to-railsolution for the amplifier must be used. This would make it more complexand probably reduce its bandwidth.

8.2 Future Improvements

The input transistors of the OTA should perhaps be PMOS instead of NMOS[4]. The output capacitance would then most certainly decrease and lead tohigher unity gain frequency and higher phase margin, since the folding pointis moved down to the NMOS transistors, which are smaller. This wouldreduce the bias current needed and thereby reduce the power. On the otherhand, the gain would decrease, but that has shown to be quite easy to takecare of with gain boosting.

Another good thing with using PMOS transistors at the input is that thecommon mode level would have to be decreased. The overall common modelevel would then decrease and make it possible to use NMOS transistors asswitches instead of transmission-gates, since the NMOS is more suitable forlower input voltages.

The second option is to decrease the sampling capacitor, thereby decreasingthe power consumption. The signal swing would then have to be increased tomaintain the SNR, thereby forcing to use bootstrapped switches. A rail-to-rail solution for the amplifier would most likely also be needed, but thatsolution might make it hard to use gain boosting due to potential instabilityof the amplifier.

APPENDIX I – DERIVATION OF TOTAL OUTPUT NOISE

76


Figure I-0-1 Track-and-hold with parasitic CGS included.

In these derivations only the CTk B -noise and noise from the input

transistors of the amplifier are considered, it is also assumed that the inputsignal is a sine wave. More details can be found in [15].

The thermal noise from the input transistors can be modelled as a voltage

source on the input of magnitudem

Bg3

Tk8(rms).

The capacitors are not generating any noise (ideally), but they sample the

thermal noise from the surroundings, which give rise to the CTkB -noise.

They are also modelled as a voltage source, but of magnitudeS

BC

Tk(rms).


77

Figure I-2 Model of input noise (mean square values and C=CS, k=kB) [4].

To calculate the total equivalent noise at the input the following method wasapplied.

1. Short-circuit the input.

2. Calculate the total noise on the output.

3. Transfer this noise back to the input.

The total output noise became:

+=

m

B

S

B2nOUT g3

Tk8

C

Tk2v

Since the gain of the amplifier in this T&H (including feedback) is one, thetotal input noise becomes

+=

m

B

S

B2nIN g3

Tk8

C

Tk2v

Assuming CS=1.5pF and gm=7mS, results in a CTkB -noise that is about

9108.1 × times larger than the thermal noise. This clearly shows that thethermal noise can be ignored. The total input noise then becomes

S

B2nIN C

Tk2v =

Until now CGS has been ignored. If it is also taken into account, the totaleffective sampling capacitor, on which the noise is sampled, becomes equalto GSSTOT CCC += . If CGS is included, the total input noise becomes


78

GSS

B2nIN CC

Tk2v

+=

What is left is to use this result to derive the total output noise during thehold phase.

During the track phase the noise is sampled (saved) on GSSTOT CCC += ,the total “noise charge” on these capacitors is

( )

( )GSSB2n

GSS

B2INn,

22TOT

2n

CCTk2Q

CC

Tk2vvoltagenoisesampledtheVVCQ

+=

=+

=====

The noise voltage during the hold phase is calculated below

26.1C

Tk2

pF4.0C;pF5.1C

C

CC

C

Tk2

C

CCTk2

C

Qv

S

B

GSS

S

GSS

S

B2S

GSSB2

S

2n2

OUT,n

×≈

≈≈≈≈

≈+

=+

==

APPENDIX II – DERIVATION OF DEPENDENCY TABLE

79


The formulas that are used to create the dependency table are based onresults derived in [4] and in chapter 3.2.

The expression for the DC-gain of a folded cascode amplifier without gainboost is

( )

⇒⇒=

===∝

+

+

∝

⇒∝

∝⇒

++

=

L6D0D

16D14D12D10D

14D14

14

6

6D

16

16D

14

14D

10D10

10

12

12D

10

10D

0D0

0

0

Dds

Dm

14m

6ds16ds14ds

10m

12ds10ds

0m0

II

IIII

IL

W

L

I

L

I

L

I

IL

W

L

I

L

I

IL

W

A

IL

1g

IL

Wg

g

ggg

g

ggg

A

++

∝

6

0D

16

10D

1414

10D

1010

10D

12

10D

0D0

0

0

L

I

L

I

WL

I

WL

I

L

I

IL

W

A

The A0-column in the dependency table (Table 6-2 page 54) can be createdby inspection of the equation above. If for example the widths of the input-transistors (W0,1) are increased, inspection of the equation above clearlyshows that the gain then increases.


80

The unity gain frequency is derived below

⇒π

=π

≈≈

≈=≈

π≈ω

π=

L

0m

L

out

out

0m110TT C

g

2

1

C

g

g

g

2

1

bandwidth-3dB

polelowesttheppA

2

1

2

1f

0D0

0

LT I

L

W

C1

f ∝

The expression for the lowest pole (p1), which is also approximately equal tothe 3dB-bandwidth, is derived below. gout is the total output admittance andCL is the total load capacitance on the output.

( )⇒

++=∝

14m

6ds16ds14ds

10m

12ds10ds

LL

out1 g

ggg

g

gg

C

1

C

gp

++∝

6

0D

16

10D

1414

10D

1010

10D

12

10D1 L

I

L

I

WL

I

WL

I

L

Ip

The expression for p2 (the second pole) is finally derived. CP in the equationbelow is equal to the parasitic capacitances at the folding point (A in Figure5-2 page 41).

0db14sb14sg16dg6dg16db6dbP CCCCCCCC ++++++=

Csg14 and Csb14 are dominating, CP is therefore approximated with

r)transistotheofareathetwice(LWLWCCC 1414141414sb14sgP =+∝+≈

⇒≈≈ 10D14

14

1414P

14m2 I

L

W

LW2

1

C

gp

141414

10D2

WLL

Ip ∝

REFERENCES

81

REFERENCES

[1] Product Specification, “10-Bit 80MSPS Sampling Analog-to-DigitalConverter”, Nordic VLSI ASA, Aug. 2001.

[2] Jiren Yuan, “Mixed Mode IC Design-Lecture notes”, Lund University,March 2001.

[3] Torkel Glad, Lennart Ljung, “Reglerteknik – Grundläggande teori”,Studentlitteratur”, 2nd edition 2000, ISBN 91-44-17892-1.

[4] David Johns, Ken Martin, “Analog integrated circuit design”, JohnWiley & Sons”, 1997, ISBN 0-471-14448-7.

[5] M. Hasan, D. R. Allee, M. Kabir and K. Rahman, “A 15-bit CMOSCyclic A/D Converter with Correlated Double Sampling”, IEEECircuits and Systems, Vol. 1, pp. 26-29, May 2002.

[6] R. Gregorian and G. C. Temes, “Analog MOS Integrated Circuits forSignal Processing”, Wiley – Interscience, 1986, ISBN 0-471-09797-7.

[7] Ben G. Streetman, Sanjay Banerjee, “Solid State Electronic Devices”,Prentice Hall, 5th edition 2000, ISBN 0-13-025538-6.

[8] Thomas H. Lee, “The Design of CMOS Radio-Frequency IntegratedCircuits”, Cambridge University Press, 2001, ISBN 0-521-63922-0.

[9] Behzad Razavi, “Design of Analog CMOS Integrated Circuits”,McGraw-Hill Higher Education, 2001, ISBN 0-07-238032-2.

[10] M. Duppils, J-E Eklund and C. Svensson, ”A study of the non-idealproperties of sample-and-hold circuits with respect to the analogbandwidth”, Advanced A/D and D/A Conversion Techniques and theirApplications, Conference Publication No. 466, July 1999.

[11] W. Yu, S. Sen and B. H. Leung, “Distortion Analysis of MOS Track-and-Hold Sampling Mixers Using Time-Varying Volterra Series”,IEEE Transactions on circuits and systems-II: Analog and Digital SignalProcessing, vol. 46, No. 2, Feb. 1999.

[12] Mohamed Zin, Kobayashi, Ichimure, San, Onaya, Kimura, Yuminaka,Sasaki, Tanaka and Abe, “A High-Speed CMOS Track/Hold Circuit”,IEEE Circuits and Systems, Vol. 3, pp. 1709-1712, Sept. 1999.

REFERENCES

82

[13] Behzad Razavi, “Design of Sample-and-Hold Amplifiers for High-Speed Low-Voltage A/D Converters”, IEEE Custom Integrated CircuitsConference, pp. 59-66, May 1997.

[14] S. Brigati, F. Maloberti and G. Torelli, ”A CMOS Sample&Hold forHigh-Speed ADC’s”, IEEE Circuits and Systems, Vol. 1, pp. 163-166,1996.

[15] Thomas Cho, “Low-Power Low-Voltage Analog-to-Digital ConversionTechniques using Pipelined Architectures”, UC Berkley PhD Thesis,1995.

[16] A. Boni, A. Pierazzi and C. Morandi, “A 10-b 185-MS/s Track-and-Hold in 0.35µm CMOS”, IEEE J. Solid-State Circuits, Vol. 36, pp. 195-203, Feb. 2001.

[17] A. R. Bugeja, B-S Song, P. L. Rakers and S. F. Gillig, “A 14-b, 100-MS/sCMOS DAC Designed for Spectral Performance”, IEEE J. Solid-StateCircuits, vol. 34, No. 12, December 1999.

[18] D. Seo, A. Weil and M. Feng, “A Very Wide-Band 14Bit, 1GS/s Track-and-Hold Amplifier”, IEEE Circuits and Systems, Vol. 5, pp. 549-552,May 2000.

[19] Mani Soma, “Fault Modeling and Testing Generation for Sample-and-Hold Circuits”, IEEE Circuits and Systems, Vol. 4, pp. 2072-2075, June1991.

[20] B. Hernes, O. Moldsvor and T. Saether, “A –80dB HD3 Opamp in 3.3VCMOS Technology using Tail Current Compensation”, IEEE Circuitsand Systems, Vol. 1, pp. 216-219, May 2001.

[21] A. Boni and A. Pierazzi, “A 150 MHz Track-and-Hold Amplifier in0.35µm CMOS”, Advanced A/D and D/A Conversion Techniques andtheir Applications, pp. 90-93, July 1999.

[22] S. Karthikeyan, S. Martezapour, A. Tammineedi and Edward K. F. Lee,“Low-Voltage Analog Circuit Design Based on Biased InvertingOpamp Configuration”, IEEE Transactions on circuits and systems-II:Analog and Digital Signal Processing, vol. 47, No. 3, March 2000.

[23] Y-I Park, S. Karthikeyan, F. Tsay and E. Bartolome, “A 10b100MSamples/s CMOS Pipelined ADC with 1.8V Power Supply”,IEEE Solid-State Circuits Conference, pp. 130-131, Feb. 2001.

REFERENCES

83

[24] Y-I Park, S. Karthikeyan, F. Tsay and E. Bartolome, “A low power 10bit, 80MS/s CMOS pipelined ADC at 1.8V power supply”, IEEECircuits and Systems, Vol. 1, pp. 580-583, May 2001.

[25] L. Sumanen, M. Waltari and Kari A. I. Halonen, ”A 10-bit 200-MS/sCMOS Parallel Pipeline A/D Converter”, IEEE J. Solid-State Circuits,Vol. 36, No. 7, July 2001.

[26] Course Material, “Chapter 14, Analog memories and related circuits”,DESC, 1991.

[27] U. Moon, G. Temes, E. Bidari, M. Keskin, L. Wu, J. Steensgaard and F.Maloberti, “Switched-Capacitor Circuit Techniques in SubmicronLow-Voltage CMOS”, International Conference on VLSI and CAD, pp.349-358, Oct. 1999.

[28] C. Eichenberger and W. Guggenbuhl, “On Charge Injection in AnalogMOS Switches and Dummy Switch Compensation Techniques”, IEEET. Circuits and Systems, Vol. 37, No. 2, Feb. 1990.

[29] Yunteng Huang, Gabor C. Temes and Paul F. Ferguson, “Novel High-Frequency Track-and-Hold Stages with Offset and GainCompensation”, IEEE Circuits and Systems, Vol. 1, pp. 155-158, 1996.

[30] Yunteng Huang, Gabor C. Temes and Paul F. Ferguson, Jr., “Offset-andGain-Compensated Track-and-Hold Stages”, IEEE Electronics,Circuits and Systems, Vol. 2, pp. 13-16, Sept. 1998.

[31] P. Yu, S. Shehata, A. Joharapurkar, P. Chugh, A. Bugeja, X. Du, S-U.Kwak, Y. Papantonopoulos and T. Kuyel, “A 14b 40Msamples/sPipelined ADC with DFCA”, IEEE Solid-State Circuits Conference, pp.136-137, Feb 2001.

[32] U. Gatti and F. Maloberti, “Fully Differential Switched-CurrentBuilding Blocks”, IEEE Circuits and Systems, Vol. 3, pp. 1400-1403,May 1992.

[33] P. R. Gray, P. J. Hurst, S. H. Lewis and R. G. Meyer, “Analysis andDesign of Analog Integrated Circuits”, John Wiley & Sons Inc., 4th

edition, 2001, ISBN 0-471-32168-0.

[34] Bsim3v3 Manual

REFERENCES

84

[35] C. Nordling, J. Österman, “Physics Handbook”, Studentlitteratur, 5th

edition 1996, ISBN 91-44-16575-7.

På svenska

Detta dokument hålls tillgängligt på Internet – eller dess framtida ersättare –under en längre tid från publiceringsdatum under förutsättning att inga extra-ordinära omständigheter uppstår.

Tillgång till dokumentet innebär tillstånd för var och en att läsa, ladda ner, skrivaut enstaka kopior för enskilt bruk och att använda det oförändrat förickekommersiell forskning och för undervisning. Överföring av upphovsrätten viden senare tidpunkt kan inte upphäva detta tillstånd. All annan användning avdokumentet kräver upphovsmannens medgivande. För att garantera äktheten,säkerheten och tillgängligheten finns det lösningar av teknisk och administrativ art.

Upphovsmannens ideella rätt innefattar rätt att bli nämnd som upphovsman i denomfattning som god sed kräver vid användning av dokumentet på ovan beskrivnasätt samt skydd mot att dokumentet ändras eller presenteras i sådan form eller isådant sammanhang som är kränkande för upphovsmannens litterära ellerkonstnärliga anseende eller egenart.

För ytterligare information om Linköping University Electronic Press se förlagetshemsida http://www.ep.liu.se/

In English

The publishers will keep this document online on the Internet - or its possiblereplacement - for a considerable time from the date of publication barringexceptional circumstances.

The online availability of the document implies a permanent permission foranyone to read, to download, to print out single copies for your own use and to useit unchanged for any non-commercial research and educational purpose. Subsequenttransfers of copyright cannot revoke this permission. All other uses of the documentare conditional on the consent of the copyright owner. The publisher has takentechnical and administrative measures to assure authenticity, security andaccessibility.

According to intellectual property law the author has the right to be mentionedwhen his/her work is accessed as described above and to be protected againstinfringement.

For additional information about the Linköping University Electronic Press andits procedures for publication and for assurance of document integrity, please referto its WWW home page: http://www.ep.liu.se/

© Erik Säll

design of a low power, high performance track-and …18676/fulltext01.pdf · design of a low power,...

Documents